deps-dev(deps-dev): bump darwin-agents from 0.7.1 to 0.9.0 in the patch-and-minor group

[dependabot]

Bumps the patch-and-minor group with 1 update: darwin-agents.

Updates darwin-agents from 0.7.1 to 0.9.0

Release notes

Sourced from darwin-agents's releases.

v0.9.0 — drift-detection canary + cross-family critic check

Two additive, opt-in features. Full detail in CHANGELOG.md.

Validate-by-reproduce drift canary (Phase 2 A5) — a new darwin canary <agent> command plus exported metrics. The A/B gate guards prompt quality; the canary guards behaviour. It compares recent execution trajectories against a version-pinned baseline using tolerance-based metrics (unordered tool-set Jaccard, ordered sequence similarity, turn-count ratio, error-rate delta) to catch drift the quality score misses — a model update or a broken tool changing how an agent works while the score stays flat. Exact-hash is avoided (LLM runs are non-deterministic); drift needs a pattern (≥2 of N runs); the baseline is pinned to the active prompt version so an intentional evolution reports insufficient-data, not a false alarm. Pure and zero-dep. --json + --exit-on-drift for CI. Closes the last unbuilt Phase-2 stream (A1–A5 complete).

Cross-family critic check — warns, or hard-fails under DARWIN_REQUIRE_CROSS_FAMILY, when all critics collapse onto one model family. claude-cli and anthropic-api are the same family, so without a second provider key the LLM-as-judge diversity guarantee is silently gone.

545 tests green. Also backfills the 0.8.0 changelog entry (it shipped to npm without one).

Changelog

Sourced from darwin-agents's changelog.

[0.9.0] — 2026-06-21

Added

• Validate-by-reproduce drift-detection canary (Phase 2 A5) — new src/evolution/canary.ts + a darwin canary <agent> command. The A/B gate guards prompt quality; the canary guards behaviour. It compares an agent's recent execution trajectories (captured in A1) against a frozen baseline using tolerance-based metrics — unordered tool-set Jaccard, ordered tool-sequence similarity, turn-count ratio, error-rate delta — and flags drift the score can miss (a model update or a broken tool changing how the agent works while the quality score stays flat). Exact-hash equivalence is deliberately avoided (LLM runs are non-deterministic). Drift requires a pattern (default ≥2 of N runs), and the baseline is pinned to the active prompt version so an intentional evolution reports insufficient-data (re-baseline) rather than a false alarm. Pure and zero-dep; --json + --exit-on-drift for CI. The metrics and the runCanaryOverExperiments orchestrator are exported from the package root so consumers can run the same check on their own captured trajectories.

• Cross-family critic diversity check (src/evolution/critic-families.ts). Multi-critic evaluation spreads critics across model families to reduce LLM-as-judge bias — but only when more than one provider key is present. Otherwise all three critics collapse onto a single family (claude-cli and anthropic-api are the same family, differing only in latency) with no signal to the operator. The run path now warns when the critics share one family, and hard-fails when DARWIN_REQUIRE_CROSS_FAMILY is set (CI / strict setups). Default behaviour is unchanged apart from the new warning.

[0.8.0] — 2026-06-21

Evolution wired end-to-end through the CLI, plus three correctness fixes. The v0.7 evolution surfaces existed but the darwin evolve command couldn't fully drive them; this release closes that gap and hardens score parsing. The automatic loop is behaviour-preserving — the gated path still passes the existing loop/GEPA suites unchanged.

Added

• darwin evolve <agent> --force now runs the loop's real variant-generation + A/B-start path on demand (DarwinLoop.forceEvolve()), bypassing the automatic "enough runs / actionable patterns / data quality" gates while still refusing the impossible cases (no active prompt, no recorded experiments, an A/B test already running). It was previously a stub that printed "not yet available" while the help text advertised it.
• src/evolution/build-loop.ts — shared loop wiring (legacy optimizer + opt-in GEPA) used by both cli/run.ts and the new --force command.

Fixed

• darwin evolve <agent> --enable|--disable now persists. The flag mutated

... (truncated)

Commits

• ffc3ecb docs(readme): document the v0.9 drift canary + cross-family judging
• 1e05cba feat(v0.9.0): drift-detection canary (A5) + cross-family critic check
• 5f174ff release: darwin-agents@0.8.0 (OSS improvement wave S1372)
• ae9ffb0 Wire v0.7 evolution into the loop + CLI; fix CLI persistence & score-parsing ...
• 5098e01 benchmark: --runs N averaging + 10 tasks + budget-cap handling, with a commit...
• See full diff in compare view

Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting @dependabot rebase.

---

Dependabot commands and options

You can trigger Dependabot actions by commenting on this PR:

• @dependabot rebase will rebase this PR
• @dependabot recreate will recreate this PR, overwriting any edits that have been made to it
• @dependabot show <dependency name> ignore conditions will show all of the ignore conditions of the specified dependency
• @dependabot ignore <dependency name> major version will close this group update PR and stop Dependabot creating any more for the specific dependency's major version (unless you unignore this specific dependency's major version or upgrade to it yourself)
• @dependabot ignore <dependency name> minor version will close this group update PR and stop Dependabot creating any more for the specific dependency's minor version (unless you unignore this specific dependency's minor version or upgrade to it yourself)
• @dependabot ignore <dependency name> will close this group update PR and stop Dependabot creating any more for the specific dependency (unless you unignore this specific dependency or upgrade to it yourself)
• @dependabot unignore <dependency name> will remove all of the ignore conditions of the specified dependency
• @dependabot unignore <dependency name> <ignore condition> will remove the ignore condition of the specified dependency and ignore conditions

[dependabot]

Labels

The following labels could not be found: dependencies. Please create it before Dependabot can add it to a pull request.

Please fix the above issues or remove invalid values from dependabot.yml.